Skip to main content
Session memory is a memory space that holds what was said in each conversation. Messages accumulate as they are exchanged, and once a conversation grows long, older content is folded into a summary.

​
When to use it

  • Continuing an earlier conversation when a user speaks as if the previous exchange is a given: “we talked about that earlier”
  • Going back and forth several times when something decided earlier has to carry into later replies
  • Content that doesn’t need to last when it’s fine for the conversation to be cleaned up once it ends
For content that has to outlive the conversation, use Wiki memory or Index memory instead.

​
Conversations tab

The conversation list is on the left, and the selected conversation’s content is on the right. That content is split into two parts.
  • Rolling summary older content folded into a summary. When a summary is itself summarized again, an L1 or L2 badge shows how many times it has been folded.
  • Recent turns the most recent conversation that hasn’t been summarized yet, in chronological order.
This tab is read-only. Unlike Wiki and Index memory spaces, you can’t edit or delete memories here. Conversations are cleaned up according to the Store setting in the Settings tab.
  • redis conversations disappear when the session ends or the store restarts.
  • postgres conversations persist, so clear them by scope with nora memory.
  • both Redis and Postgres are used together. Each store behaves as described above.
Session has no Log tab, because the conversation itself is this memory space’s log.

​
How memories accumulate

  • Written automatically every time a conversation is exchanged, the user input and the Agent response accumulate in this memory space automatically.
Conversations are separated by the thread_id scope key. For how to pass threadId from a trigger, see Trigger block.